Skip to content

LucidFusion

Generate high-resolution 3D Gaussians from sparse views

Visit official site

What LucidFusion is

The framework is described as a flexible end-to-end feed-forward system for generating high-resolution 3D Gaussians from unposed and sparse multi-view images. It uses a relative coordinate map to align geometry between views.

What you can do with it

  • Feed any number of unposed views into the model
  • Align geometric features with a relative coordinate map
  • Merge with a single-image-to-3D pipeline

Example uses

  • A film character is rebuilt from angled images
  • A creature model is prepared for post-production
  • A nested doll is reconstructed for an exhibit

Pros & cons

✓ What we like

  • Cross-dataset content creation is supported
  • Detailed 3D Gaussians at 512x512 are produced

! What to watch out for

  • Input is a set of multi-view images
  • The pipeline uses a Stable Diffusion model in the forward pass

FAQ

What resolution is produced?

The features state detailed 3D Gaussians at 512x512 resolution.

How are views aligned?

A relative coordinate map aligns geometric features between views.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review